airplane

Travel AI with Pana

Travel’s an interesting industry because it’s inherently global which makes it inherently complex, and it’s so behind other industries when it comes to innovative and advanced technology being applied. A great example of that is when you buy a ticket on an Expedia or Priceline, etc., it’s likely that 75% of the time that a fax is sent to the hotel to tell them that you’ll be staying there that night.

Ginette: I’m Ginette.

Curtis: And I’m Curtis.

Ginette: And you are listening to Data Crunch.

Curtis: A podcast about how data and prediction shape our world.

Ginette: A Vault Analytics production.

Ginette: Data Crunch is brought to you by data.world, the productive, secure platform for modern data teamwork. Organizations like The Associated Press, Rare, Encast, and Square Panda use data.world to replace outdated barriers with deep connections among data, people, and impact. This makes data easier to find, helps people work together better, and puts data and insights in the hands of those who need it. To learn more, visit data.world and request a demo.

Curtis: Envision in your mind’s eye our globe and all the airplane flights in the sky at any given time. Now, zoom into a busy city on that globe and notice all the cars being rented by business professionals and the hotels that they’re checking into. Even in just one city, the amount of transactions is dizzying. The travel industry has a lot going on, and yet, sometimes it’s surprisingly antiquated.

Devon: I’m Devon Tivona. I’m a founder at Pana. My background is actually technical. I went to school for engineering, spent the first five years of my career as a engineer, then a product lead, and most recently as a founder of this company.

Ginette: The founders of Pana were intrigued with the possibilities of what they could do in the professional travel space, and as they talked with travelers, they saw an opportunity.

Devon: We were talking particularly to frequent traveler[s]. And we kept hearing over and over again two primary pain points. One was felt like “with all the new found technology in the travel space, I still have to be my own travel agent. And it was great 10 years ago when I could just email someone, and they would take care of all of the logistics for me, but now all the technology has made it so I have to do all that work.”

And then the second pain point that we started hearing was “then once I buy my plane ticket or my hotel ticket, if I need to make a change or something goes wrong and I want to get ahold of a real human being, that’s like pulling teeth from these companies, particularly if I bought my ticket online.” So we kind of had this vision for could we build the 21st century version of the travel agent, but do so, you know, in a scalable Internet business sort of way. We didn’t want to build a boutique travel agency. We wanted to build something big.

Travel’s an interesting industry because it’s inherently global which makes it inherently complex, and it’s so behind other industries when it comes to innovative and advanced technology being applied, particularly because it’s so big, not because it doesn’t have awesome people working in the space. A great example of that is when you buy a ticket on an Expedia or Priceline, etc., it’s likely that 75% of the time that a fax is sent to the hotel that you’ll be staying there that night. And for me when I heard that I was like, “okay, this is a really interesting industry because I can always be building stuff here as a technologist.”

Curtis: Pana focused on the corporate travel space in particular because it felt it had more user pain points than other travel workflows.

Devon: I think that there’s, a there’s a lot of a lot of varied user pain that are experienced throughout a travel journey, particularly I would say on the corporate travel side of things. I think that leisure travel, there’s billions of dollars being spent on optimizing conversion flows of you buying a plane ticket on Kayak, or you purchasing a hotel on Hotel Tonight, so they actually have done a pretty good job from a UI/UX for making that experience seamless. Business travel there’s been less investment there, partially because there’s less incentive to invest in user interfaces in a B2B product thank a B2C product, which we’re trying to change.

Curtis: Sometimes when building a product you have to beware of the hype of new technology and really evaluate if it’s the best thing for your users. Especially when you’re talking about AI, which right now is in an extreme hype phase.

Devon: We were founding Pana along the same time that the F8 developers conference, Facebook’s developers’ conference, when they announced their chatbot platform, and along with that, there was a huge wave of hype around building text interfaces that had natural language processing behind them, and while some of the tech demos were more impressive than others, I think what we found by and large, what you can create today is really a natural language processing on the user input side and a fairly unsophisticated decision tree on the back end for you know that is essentially the text equivalent that was like “press A for flights, press B for hotels,” and when we started going down that route and investigating what technology we would want to spin up there, how real we could make the interaction feel, what the actual utility of an interaction would be, we found that trying to entirely automate the conversation and turn and try and imitate a human conversation and convince the user that they were talking to a human even though they were talking to a bot, really wasn’t worth the ROI that could be gained there.

Ginette: While vetting this business concept, Devon’s team conducted an interesting study to understand the smallest details of the corporate travel agency world.

Devon: We went to a traditional travel agency, a corporate travel agency,  and we sat with their agents, and it was almost like anthropological in our study, but we would time the amount of time that it would take for these agents to do various activities, so like even like they wrote an email to Bob about a flight change. How long did that email take? They booked a ticket online using the system. How much did that take?

And we basically stack ranked, you know, where was the time being spent for a travel agent today? And surprisingly we came to the conclusion that actual communication with customers, the back and forth of writing emails etc., wasn’t where travel agents were spending the majority of their time. And in fact, they were spending a majority of their time wrangling the other data sets that a travel agent has to deal with, whether it be making a flight booking, making a flight change, cancelling a hotel, extending a stay, picking out options that match a user’s preferences, making sure that the invoicing system was set up correctly so that everything flowed to the accounting team. Those types of nitty gritty very data movement travel-related tasks were the ones that were taking up time, not communicating with customers.

Curtis: After realizing that it wasn’t communication with customers that was taking up the majority of the time, Pana focused on improving the interaction with the datasets travel agents have to interface with behind the scenes.

Devon: Instead the approach we decided to take was let’s actually have humans power our service, at least for the next five years, and let’s instead of trying to replace them through an AI, let’s build a really, really powerful agent dashboard that they can sit in front of in order for them to do their job efficiently and the best that they can, and it’s turned out that that investment in that product has been one of the best strategic decisions we’ve made because on the front end, we get the benefit of a real human conversation. If you want a Pana agent to tell you a joke, they may not be good at one but they would reply and tell you a joke, but on the backend, we get to continually push improvements to the agent workflows and to the agent dashboard that, that over and over again increase agent efficiency, decrease our costs, and increase the number of things we can do for our customers, and that’s really where we continue to innovate and where we continue to evolve.

And it’s never one algorithm or one approach that solves the problem. It’s really an orchestrated approach of hundreds of various systems and integrations and techniques being run in the background that makes everything come together.

Curtis: Pana has very specific insights into their workflows, and they make work easier for their travel agents.

Devon: We’re reaching a local maxima of how better you can display the eight basic data points you need to make a flight purchase, and the 14 different data points you need to make a hotel purchase.

Another element that that’s integrated in is a messaging platform, so agents don’t need to switch over from a search platform to a messaging platform to talk to our customers or send over flight or hotel options. And there’s some data science being applied there, so we are doing some high-level natural language processing, not to generate responses or to fake that it’s a human talking to you when it’s actually a bot, but actually to pull out key entities, like airports, times with the right time zones, when you say next week, what do you mean by that? And pulling those out and tokenizing those in a way that’s really easy for agents to copy information out of that track conversation and avoid transcription error.

Ginette: As a user, on a front end of the product, you may simply experience a smooth process, but the backend is being constantly probed for improve efficiencies.

Devon: We actually have a really really big complex to do list that the agents work off of. So at any given time, very similar to that exercise that I mentioned in the beginning where we were stopwatch timing different categories of work that traditional travel agents did, our product does that automatically for our travel agents that sit in front of our platform, so we know exactly how long it takes on average along with the standard deviation of how long a flight search takes. We know exactly how long a hotel search takes, we know exactly how long an amendment takes, and that gives us a stack ranked list of things to work on and things to optimize that give us highest ROI for our business.

One of the interesting insights that we’ve gained is that it’s a really complex problem set to optimize around the hundreds of potential workflows that you could have in a travel setting, so we’ve started looking at use case specific workflows and seeing if we can become world class at those workflows. So one of those that we’ve already launched publicly was a recruiting workflow, so if you are a company who is hiring someone, and you’re bringing them in for an on-site interview, typically the way that that’s done is your recruiting coordinator is going to play travel agent and send options back and forth with the candidate and book them in using company card and stuff like that, so we released a product that does just that for a company, so if they want to buy just Pana for their recruiters or their talent team, they can. And we are world class at that workflow. There’s no one that handles recruiting travel better, and by doing that and by verticalizing some of our products, we’ve found that we can grow a lot quicker, so we started in recruiting and some other verticals we’re looking at, event travel being one of those, so if you’re hosting a big conference for your company, that’s one space we’re looking at as well.

Curtis: In addition to working on specific travel verticals, Pana also personalizes the search process.

Devon: We collect over 400 data points about our customer that’s both learned and inferred over usage of the product with a little bit of a cold start survey at the beginning. And then those preferences and data points about the customer surface throughout the interface at the right place, so when I’m booking you a flight, I know that you like staying upfront in bulkhead seats in the aisle, and that information is contextually provided to me when I’m doing that workflow.

A search on customer A’s profile might look completely different than a search on customer B’s profile because their preferences are different. So we’re intelligently ranking those options that are showing up for the agents so that we’re limiting the number of subjective decisions they have to make on behalf of the customer.

The biggest problem that we’ve uncovered from a product perspective in travel is doing support right, and being able to when there’s a snowstorm in New York and an A380 and a 747 has been grounded, and you have 40 travelers across both of those, how do you scalably, efficiently, and with some sort of humanity service something like that. Or even a simple request as me saying ,”I want to extend my San Francisco trip one more day.” I think on a traditional travel website, the user’s going to be making three different phone calls and clicking around on five different websites. We at Pana have set out to solve this problem by literally can you just say, I want to change my trip and extend it an extra day.

Ginette: Like any product improvement process, it takes time to iterate into better versions, and Pana is taking it one step at a time.

Devon: I’m cautiously optimistic about some really, really innovative technologies that can make it possible for Pana to deliver our service worldwide at a significantly lower cost than our competitors can today, and that’s really what it’s all about. It’s about delivering better experiences at a lower cost. But I think we’re going to be cautiously optimistic is the right word because we’re never going to employ efficiency technology that means it’s going to decrease the user experience, so you’d never going to see a human interaction getting replaced by a technology interaction if we think that that interaction is going to be worse in any way.

Ginette: A huge thanks to Devon Tavona from Pana for speaking with us, and as always, check out datacrunchpodcast.com for show notes and attribution. And if you’d like to easily learn about the latest artificial intelligence and machine learning applications, go sign up for our weekly newsletter at datacrunchpodcast.com/ai.

Sources

Music source

Picture source